Challenges of Network
Optimization in a
WAN-Cloud World
@AtchisonFrazer
April 2017
2
Large global manufacturer (automotive):
Core challenge: Overinvesting in
‘Bandwidth Capacity Planning’
“Our network does not represent a steady
state environment. It is constantly having
new demands placed on it; each of these
demands require more bandwidth to be
always available. If that bandwidth is not
available, the network must be able to buffer
traffic sufficiently until the congestion clears.”
“Make My Network Flow Like Air”
3
• Pause + Rate Limit
• Buffering + Random Dropping
• Congestion Diagnostics – L3
L3L2
Network Disruption
15%
Performance
15%
Performance
Software Analysis
Congestion 15%
Performance
Packet Drop/Loss
15%
Performance
Buffering
CONGESTION MITIGATION
• Random path routing
• Rogue broadcasting (NIC)
• Last Mile edge penalty
CORE NETWORKING
Traditional L2-L3 Networking Limitations
4
Conventional Assumptions: How to Avoid . . .
Premium
Vendor Lock-in
Installed
Base
Lifecycle
Churn
Optimization
Software &
Services
Layers
Without visibility to QoE application behavior and associated packet bandwidth, organizations
are either likely to over-provision hardware assets or connectivity connections, leaving too
little headroom for conventional Cloud-WAN networks to scale with end-user demands.
5
High Touch Engineer
Core Infrastructure (L2/L3)
Network Analytics Software
Break/Fix Maintenance with Smart Diagnostics
Professional Consulting
Application Visibility
Architecture Advisory
Trap #1: Legacy Equipment Optimization Tax
6
Server A
Server B
Server C
Server D
Server E
Server F
S0
L0 L3
S1 S2 S3
L1 L2 L4 Ln
• Server A is sending a lot of frames to server D
• Server B becomes active sending frames to server E; server C becomes active sending frames to server F
• Links between node L0 and S0 and S0 and L3 congest; WAN Last Mile penalized
Trap #2: Traditional Static Network Orthodoxy
7
Congestion
15%
Performance
15%
Performance
15%
Performance
15%
Performance
T1
MPLS
• Random path routing
• Rogue broadcasting (NIC)
• Last Mile edge penalty
CORE NETWORKING
Trap #3: Over-Provisioning Hardware & Bandwidth
8
BRANCHDATACENTER
MPLS
SAAS
Internet
Link quality measured for EVERY packet of EVERY application continually in EACH direction?
Allocates available bandwidth dynamically end-to-end based on predictive analytics
for degrading conditions and failsafe maneuvers for congestion avoidance?
MAC
IP
TCP/UDP
DATA
Packet
Millisecond
measurement of
latency, loss, jitter
probabilities?
What happens when packets reach the BRANCH?
Typical WAN Branch Network Topology
App-aware: Real-Time?
Interactive? Bulk?
Performance Control?
Cloud-aware: Resilient
Connection? Disaster
Recovery? QoS Metrics?
QoS b/w slack
for business
critical apps with
QoE metrics?
9
Talari Networks Next Gen SD-WAN: Cloud/App-Aware Optimization
What happens in a Talari powered BRANCH?
10
@AtchisonFrazer
Talari.com

Challenges of Network Optimization in a WAN-Cloud World

  • 1.
    Challenges of Network Optimizationin a WAN-Cloud World @AtchisonFrazer April 2017
  • 2.
    2 Large global manufacturer(automotive): Core challenge: Overinvesting in ‘Bandwidth Capacity Planning’ “Our network does not represent a steady state environment. It is constantly having new demands placed on it; each of these demands require more bandwidth to be always available. If that bandwidth is not available, the network must be able to buffer traffic sufficiently until the congestion clears.” “Make My Network Flow Like Air”
  • 3.
    3 • Pause +Rate Limit • Buffering + Random Dropping • Congestion Diagnostics – L3 L3L2 Network Disruption 15% Performance 15% Performance Software Analysis Congestion 15% Performance Packet Drop/Loss 15% Performance Buffering CONGESTION MITIGATION • Random path routing • Rogue broadcasting (NIC) • Last Mile edge penalty CORE NETWORKING Traditional L2-L3 Networking Limitations
  • 4.
    4 Conventional Assumptions: Howto Avoid . . . Premium Vendor Lock-in Installed Base Lifecycle Churn Optimization Software & Services Layers Without visibility to QoE application behavior and associated packet bandwidth, organizations are either likely to over-provision hardware assets or connectivity connections, leaving too little headroom for conventional Cloud-WAN networks to scale with end-user demands.
  • 5.
    5 High Touch Engineer CoreInfrastructure (L2/L3) Network Analytics Software Break/Fix Maintenance with Smart Diagnostics Professional Consulting Application Visibility Architecture Advisory Trap #1: Legacy Equipment Optimization Tax
  • 6.
    6 Server A Server B ServerC Server D Server E Server F S0 L0 L3 S1 S2 S3 L1 L2 L4 Ln • Server A is sending a lot of frames to server D • Server B becomes active sending frames to server E; server C becomes active sending frames to server F • Links between node L0 and S0 and S0 and L3 congest; WAN Last Mile penalized Trap #2: Traditional Static Network Orthodoxy
  • 7.
    7 Congestion 15% Performance 15% Performance 15% Performance 15% Performance T1 MPLS • Random pathrouting • Rogue broadcasting (NIC) • Last Mile edge penalty CORE NETWORKING Trap #3: Over-Provisioning Hardware & Bandwidth
  • 8.
    8 BRANCHDATACENTER MPLS SAAS Internet Link quality measuredfor EVERY packet of EVERY application continually in EACH direction? Allocates available bandwidth dynamically end-to-end based on predictive analytics for degrading conditions and failsafe maneuvers for congestion avoidance? MAC IP TCP/UDP DATA Packet Millisecond measurement of latency, loss, jitter probabilities? What happens when packets reach the BRANCH? Typical WAN Branch Network Topology App-aware: Real-Time? Interactive? Bulk? Performance Control? Cloud-aware: Resilient Connection? Disaster Recovery? QoS Metrics? QoS b/w slack for business critical apps with QoE metrics?
  • 9.
    9 Talari Networks NextGen SD-WAN: Cloud/App-Aware Optimization What happens in a Talari powered BRANCH?
  • 10.